Cloudflare launches Clef decision-making models

Cloudflare has announced the release of Clef and Clef-flash, two new decision-making models designed to handle classification and routing tasks. Unlike general-purpose chatbots, these models evaluate predefined answers against a provided schema rather than generating free-form text. This approach allows for faster processing by bypassing token-by-token generation.

Clef is built on the 27-billion parameter Qwen3.8-27B backbone and is optimized for high accuracy. Clef-flash utilizes the 9-billion parameter Qwen3.5-9B model, prioritizing low latency and speed. Both models support multimodal inputs, including text, images, and video, and are available under the Apache License 2.0, allowing for local execution via downloaded weights.

Cloudflare provides these models through its Workers AI inference service. While Clef offers a larger context window of 65,536 tokens, its input pricing is set at $0.24 per million tokens, which is higher than some existing alternatives like Jev. The company also plans to offer fine-tuning services to help customers tailor the models to specific business logic, such as support ticket prioritization or content moderation.